Education, Science, Technology, Innovation and Life
Open Access
Sign In

Improved winter road condition classification method for autonomous driving based on hierarchical transformer

Download as PDF

DOI: 10.23977/acss.2024.080620 | Downloads: 14 | Views: 294

Author(s)

Zhengguang Lu 1, Huaqi Zhao 1

Affiliation(s)

1 School of Information and Electronic Technology, Jiamusi University, Jiamusi, China

Corresponding Author

Huaqi Zhao

ABSTRACT

Aiming at the problem that the existing road surface classification methods cannot accurately identify the conditions of winter road surface with low discrimination, we propose an improved winter road condition classification method based on hierarchical Transformer. Firstly, dynamic overlapping patch embedding is introduced, which can flexibly handle input features of any size and preserve the continuity of local detail information through dynamic position encoding and overlapping patch embedding. Then, frequency-based factorized attention is used to extract frequency features containing high-level context information, enhancing the feature representation between image categories. Finally, a multi-level feature fusion method based on weight average strategy is proposed, by evenly allocating the dynamically upgradable weights to the output layers of each stage and performing multi-level fusion, the low-level features are projected to the high-level to continue learning, the feature representation is enriched, and the discrimination of the classification image is increased, thereby the classification performance is improved. Experiments are carried out on the WRF dataset of snow and ice road. The classification accuracy of the proposed method can reach 88.93% with only 3.8M parameters and 0.6G computational complexity, which is better than the current mainstream road classification methods.

KEYWORDS

Transformer, Classification, Dynamic overlapping patch embedding, Frequency-based factorized attention, Multi-level feature fusion

CITE THIS PAPER

Zhengguang Lu, Huaqi Zhao, Improved winter road condition classification method for autonomous driving based on hierarchical transformer. Advances in Computer, Signals and Systems (2024) Vol. 8: 159-166. DOI: http://dx.doi.org/10.23977/acss.2024.080620.

REFERENCES

[1] Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017). ImageNet classification with deep convolutional neural networks. Communications of the ACM, 60(6), 84-90.
[2] Howard, A., Sandler, M., Chu, G., Chen, L. C., Chen, B., Tan, M., ... & Adam, H. (2019). Searching for mobilenetv3. In Proceedings of the IEEE/CVF international conference on computer vision (pp. 1314-1324).
[3] Dosovitskiy, A. (2020). An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929.
[4] Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., ... & Guo, B. (2021). Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision (pp. 10012-10022).
[5] Wang, W., Xie, E., Li, X., Fan, D. P., Song, K., Liang, D., ... & Shao, L. (2022). Pvt v2: Improved baselines with pyramid vision transformer. Computational Visual Media, 8(3), 415-424.
[6] Li, K., Wang, Y., Zhang, J., Gao, P., Song, G., Liu, Y., ... & Qiao, Y. (2023). Uniformer: Unifying convolution and self-attention for visual recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(10), 12581-12600.
[7] Dong, B., Wang, P., & Wang, F. (2023, June). Head-free lightweight semantic segmentation with linear transformer. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 37, No. 1, pp. 516-524).
[8] Yu, W., Luo, M., Zhou, P., Si, C., Zhou, Y., Wang, X., ... & Yan, S. (2022). Metaformer is actually what you need for vision. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 10819-10829). 

Downloads: 27683
Visits: 469053

Sponsors, Associates, and Links


All published work is licensed under a Creative Commons Attribution 4.0 International License.

Copyright © 2016 - 2031 Clausius Scientific Press Inc. All Rights Reserved.